Perception & Psychophysics 1979, Vol 25 (4), 292-302 Contextual effects in vowel perception I: Anchor-induced contrast effects JAMES R SAWUSCH and HOWARD C NUSBAUM State University of New York at Buffalo, Buffalo, New York 14226 Results from recent experiments using a selective adaption paradigm with vowels have been interpreted as the result of the fatigue of a set of feature detectors These results could also be interpreted, however, as resulting from changes in auditory memory fauditory contrast} or changing response criteria tresponse bias} In the present studies, subjects listened to vowels under two conditions: an equiprobable control, with each of the stimuli occurring equally often, and an anchoring condition, with one vowel occurring more often than any of the others Contrast effects were found in that vowel category boundaries tended to shift toward the category of the anchor, relative to the equiprobable control Results from these experiments were highly similar to previous selective adaptation results and suggest that neither feature detector fatigue nor response criterion changes can adequately account for the adaptation/ anchoring results found with vowels Theories of speech perception have proposed a number of diverse procedures for recovering invariant phonetic information from the continuously varying speech signal Among these have been theories proposing a direct link between production and perception such as motor theory (Liberman, Cooper, Harris, & MacNeilage, 1962; Liberman, Cooper, Harris, MacNeilage, & Studdert-Kennedy, 1967) and analysis by synthesis (Stevens & Halle, 1967) and more passively oriented stage theories (Fant, 1967) A number of recent theories have been proposed within an information processing framework in which information in the speech signal is transformed and recoded by a number of stages of processing (Cutting, 1976; Pisoni, 1975b; Pisoni & Sawusch, 1975; Sawusch, 1977a, Tartter & Eimas, 1975) Within these information procession models, two questions can be addressed The first asks what the levels of processing between input (speech waveform) and output (phonetic code) are, and the second asks what operations occur.within each level As one possible answer to the second question, feature detectors have recently received a great deal of attention in speech perception The primary evidence supporting the role of feature detectors in speech perception has come from studies using the selective adaptation paradigm (see Ades, 1976; Eimas & Miller, 1978) Briefly, this paradigm involves preThe research reported here was supported by SUNY Research Foundation and University Funds grants The authors would like to thank Dr David B Pisoni for making the facilities at Indiana University available for generating the audiotapes and Eileen Schwab for her assistance in running subjects and thoughtful comments on an earlier draft Reprint requests should be sent to the first author at the Department of Psychology, 4230 Ridge Lea Road, Buffalo, New York 14226 Copyright © 1979 Psychonomic Society, Inc senting the subject with repeated occurrences of an adapting sound and then testing the subject’s identification (or discrimination) of a set of stimuli These results are then compared to the same subject’s unadapted, baseline identification (discrimination) of the same stimuli The typical result has been that the category boundary for the test series shifts following adaptation Figure summarizes the type of effect that is typically found The solid line represents the subject’s rating function for the stimulus set before adaptation (baseline) When Stimulus (arrow on left) is used as an adaptor, the rating function shifts toward the Stimulus end of the series (function on left) Similarly, when Stimulus is used as an adaptor, the function shifts toward Stimulus (on the right) Ld (9 n,- 1234567 STIMULUS VALUE Figure Typical baseline (solid), post Stimulus adaptalion (open triangles) and post Stimulus adaptation (open circles) rating functions 292 0031-5117/79/040292-11501 ~35/0 CONTEXTUAL EFFECTS IN VOWEL PERCEPTION This type of result has been found for stopconsonant stimuli varying along the phonetic features of voicing (Eimas, Cooper, & Corbit, 1973; Eimas & Corbit, 1973), place of articulation (Bailey, 1975; Cooper, 1974; Diehl, 1975; Sawusch, 1977a; Tartter & Eimas, 1975), and manner (Bailey, 1975; Cooper, Ebert, & Cole, 1976; Diehl, 1976) These adaptation results have generally been interpreted as supporting the existence of feature detectors for certain auditory components in the speech signal (see Ades, 1976; Cooper, 1975; Eimas & Miller, 1978, for reviews) These feature detectors may respond to various spectral or temporal patterns in the speech signal (see Blumstein, Stevens, & Nigro, 1977; Diehl, 1976; Pisoni & Tash, 1975; Sawusch, 1977a, 1977b) Besides feature detector fatigue, two other interpretations of category boundary shifts in selective adaptation experiments have emerged Simon and Studdert-Kennedy (1978) have proposed an auditory contrast explanation in which an adaptor establishes an "auditory ground" that subsequent stimuli are compared with Their interpretation is based on a comparison of selective adaptation and anchoring procedures In their anchoring experiments, category boundaries for various sets of stimuli were compared across two conditions: an equiprobable control and an anchoring series in which one (endpoint) stimulus occured more often than the other stimuli Simon and Studdert-Kennedy found that the results of anchoring and adaptation procedures were often very similar, in that the category boundary for a set of stimuli would shift toward the more frequently occurring category (adaptor/anchor) Simon and Studdert-Kennedy (1978) concluded that feature detector fatigue was not a sufficient explanation for the adaptation and anchoring results, and proposed the auditory contrast explanation (but see Sawusch & Pisoni, Note 1, for an alternative interpretation of speech anchoring results) This interpretation is also consistent with a number of other experiments that have shown contextual influences in speech perception (Broadbent & Ladefoged, 1960; Eimas, 1963; Fry, Abramson, Eimas, & Liberman, 1962; see Simon & StuddertKennedy, 1978, for review) Recent experiments reported by Brady and Darwin (1978), Diehl, Elman, and McCusker (1978), and Elman (Note 2) are also consistent with the auditory contrast explanation Although Diehl et al (1978) have referred to their explanation as "response contrast," they have also referred to this response contrast as occurring at an auditory level of processing (similar to Simon & StuddertKennedy, 1978) Consequently, the term "auditory contrast" seems to be more appropriate The auditory contrast explanation outlined above, is, in many respects, similar to adaptation level theory as proposed by Helson (1964) In adaptation level theory, previous stimuli would be combined to form 293 an adaptation level against which new stimuli would be compared If the assumption is made that the adaptation level occurs at an auditory level of processing, then an adaptation level theory explanation of selective adaptation and anchoring results is very similar to the auditory contrast explanation just described (but see Simon & Studdert-Kennedy, 1978, for an alternative view) Since both adaptation level theory and the auditory contrast explanation interpret changes in phonetic category boundaries as due to changes in a perceptual referent, these two explanations will be grouped toegether here under the term "auditory contrast." The third interpretation that has emerged for selective adaptation results is that of a response bias on the part of subjects (Rosen, Note 3) Subjects, having heard an adapting syllable repeated many times, may simply use categories other than that of the adaptor in responding In particular, Rosen has interpreted the results of a number of anchoring experiments as supporting a range-frequency theory (Parducci, 1963, 1965, 1975)explanation of selective adaptation effects Although range-frequency theory ,was originally proposed to describe the mapping of arbitrary categories onto equally discriminable stimuli, it may be extendable to the nonarbitrary categories and the unequally discriminable sounds of speech However, for our purposes, the most important aspect about range-frequency theory is that it claims that changes in identification functions are the result of modifications in labeling strategies (response bias) and are not due to changes in a perceptual referent Thus, the range-frequency theory interpretation of the category boundary shifts in adaptation/anchoring experiments is in direct conflict with both the feature detector fatigue and auditory contrast explanations with regard to the level(s) of processing involved (See also Simon & Studdert-Kennedy, 1978; Sawusch & Pisoni, Note 1, for other evidence and views on the efficacy of a range-frequency theory interpretation of adaptation/anchoring results.) Current evidence seems to favor the feature detector fatigue and auditory contrast explanations of selective adaptation and anchoring results with stop consonants (see Eimas & Miller, 1978; Simon & StuddertKennedy, 1978; Sawusch & Pisoni, Note 1) However, all three alternatives need to be considered with regard to selective adaptation results that have been reported for a set of isolated, steady-state vowels (Morse, Kass, & Turkienicz, 1976) In their experiments, Morse et al used a set of 13 synthetic vowels which varied from [i] (as in beet) through [I] (as in bit) to [r] (as in bet) Stimuli ([i]), ([I]), and 13 ([r]) were used as adaptors However, the results of the Morse et al experiment differed from the usual findings of experiments using stop consonants In particular, two aspects of their results seem to be 294 SAWUSCH AND NUSBAUM crucial One is that Stimulus 7, the [I] adaptor, had no discrimination results, Pisoni, 1973, 1975a), then consistent effects on subjects’ responses to the test vowel identification would show contextual effects series This is one of the few times that a speech Comparison of the test items with PAS might accenadaptor, drawn from a syllable test series, had little tuate the differences between the test item and the adaptor This would lead to the classification of effect on the test series The second aspect of Morse et al.’s results that is not consistent with stop conso- vowels that are relatively ambiguous into categories nant results is that the [i] adaptor (and the [~] further away from the adaptor (a contrastive change) adaptor) had an effect at both the [i]-[I] category This explanation hinges, of course, on whether or not boundary and the [I]-[~] category boundary Cooper vowels show contrast effects (change of ambiguous (1974) had found that for a [ba]-[da]-[ga] place series, vowels into categories other than that of the context) an adapting syllable affected only the adjacent cateContextual information has been shown tc, be gory boundary (See also Foreit, 1977, for similar re- tremendously influential in vowel perception Ei, mas sults using Thai speakers and a series of stops varying (1963) and Fry, Abramson, Eimas, and Liberman along the voicing dimension.) Morse et al (1976) (1962) have shown that vowels indeed exhibit coninterpreted their vowel adaptation results as resulting trast effects These investigators had subjects identify from the fatigue of a set of relative feature detectors vowels that were presented in triads Their results However, given the differences between the vowel showed that the identification of any particular vowel adaptation results of Morse et al and previous adap- tended ~o migrate toward categories other than that tation studies with stop consonants and other well- of the vowels with which it was grouped in the triad established differences between stop consonant and This effect was especially pronounced for the relativevowel perception, this conclusion may be premature ly ambiguous vowels in their test series In a similar In particular, vowel and stop consonant perception vein, Broadbent and Ladefoged (1960; Broadbent, seem to differ in two crucial respects: the role of Ladefoged, & Lawrence, 1956) have shown that auditory memory in vowel perception and the tremen- the formant frequencies of a precursor phrase will dous influence of contextual information on vowel influence the perception of an ambiguous synthetic perception word as either "bit" or "bet." The effect that they A number of studies have shown that the free recall found was also one of contrast That is, the ambigof stop consonants and vowels differs Crowder (1971, uous word was identified as belonging to the cate-1973) has shown that for immediate recall of lists of gory that was different from (contrasted with) the specsyllables which differ only in their vowels, recency, tral information in the precursor phrase Broadbent modality, and suffix effects are found That is, the and Ladefoged (1960) interpreted their results in last syllables presented are recalled better (recency terms of a modified version of adaptation level theory effect), auditory presentation of syllables yields better (see Helson, 1964) Thompson and Hollien (1970) recall than visual presentation (modality effect), and found a similar contrast effect In their experiment, presentation of a post list item interferes with recall of subjects identified the second of a pair of synthetic the items at the e.nd of the list (suff’LX effect) Crowder vowels When the second vowel was "ambiguous" (1971, 1973; also Cole, 1973) found these effects for (had formant frequencies between those of "good vowels, but found no such effects for lists of syllables examples" of various vowels), the subjects displayed a varying only the stop consonants Cole (1973) and tendency to use categories other than that of the preCrowder (1971, 1973) have explained their results in cursor vowel in identifying the second vowel (results terms of differences in auditory memory between very similar to Fry et al., 1962) Since the adaptation vowels and stop consonants Crowder has further sug- results found by Morse et al (1976) for vowels can gested that auditory information about vowels is re- also be described as a contrast effect (ambiguous tained in precategorical acoustic storage (PAS, items near the category boundary fall into the Crowder & Morton, 1969) while auditory information adjacent or contrasting category after adaptation), the about the stops is not retained in the PAS A similar same processes that underlie the results of Broadbent explanation has been offered by Fujisaki and Kawa- and Ladefoged (1960), Fry et al (1962), Eimas shima (1969, 1970) and Pisoni (1973, 1975a) for dif- (1963), and Thompson and Hollien (1970) may be references between the categorical perception of stop sponsible for the adaptation results observed for consonants and the continuous perception of vowels vowels Recently, Sawusch and Pisoni (Note 1) have used Thus, the effects observed by Morse et al (1976), in an anchoring paradigm to investigate the perception their adaptation experiment with vowels, could have of stop consonants, vowels, and other auditory stimbeen due to influences of the adaptor on auditory memuli Using a set of synthetic, steady-state vowels varyory (PAS) for their vowel test items Specifically, at ing from [i] to [I], subjects were run in two conditions the end of a sequence of the adapting vowel, PAS In the control condition, each vowel occurred equally would contain only information pertaining to the often In the anchoring conditions, one of the adapting vowel If subjects identify the test vowels end-point vowels occurred four times as often as each with reference to the information in PAS (cf ABX CONTEXTUAL EFFECTS IN VOWEL PERCEPTION 295 of the other vowels Sawusch and Pisoni (Note 1) found that the category boundary between [i] and [I] moved toward the category of the anchoring vowel, relative to the equiprobable control Simon and Studdert-Kennedy (1978) have reported a similar result for a [bae]-[d~] CV series These contrast effects are very similar to some of the category boundary shifts found by Morse et al (1976) using the selective adaptation procedure Simon and StuddertKennedy interpreted their vowel anchoring results as due to auditory contrast, as described earlier Sawusch and Pisoni attributed their vowel anchoring results to the same processes that mediate the continuous perception of vowels From the work of Crowder (1971, 1973) and Pisoni (1973; 1975a), this would seem to be changes in auditory memory In particular, each stimulus may be categorized by a comparison with auditory memory of the immediately preceding stimuli as well as long-term memory prototypes for the relevant categories Auditory memory would contain precategorical information about previous vowel stimuli, such as formant frequencies or some ratio or difference between formant frequencies (of Gerstman, 1968) In a standard identification sequence, each stimulus occurs equally often in random order Thus, in the subject’s identification of any particular stimulus, information from every stimulus has an equal probability of being present in auditory memory and available for comparison By comparison, in an adaptation or anchoring procedure, one stimulus, the adaptor/anchor, is presented more often than the other stimuli and is thus more likely to be present in auditory memory when any particular stimulus is identified Thus, the anchor contributes disproportionately to the "auditory ground" (of Simon & Studdert-Kennedy, 1978) and ambiguous, boundary stimuli now contrast with this ground and are placed in the category opposite (away from) the anchor This description is very similar to a version of adaptation level theory proposed by Restle (1978) In this version of adaptation level theory, three factors contribute to the adaptation level and thus to the judgment of a stimulus: the stimulus itself, the immediate context, and the background (usually held constant) In terms of the present description of auditory contrast, the immediate context is provided by the information in auditory memory about the immediately preceding stimuli and the background is provided by a set of internal referents or prototypes in long-term memory Thus, auditory ground is synonymous with adaptation level, and this conceptualization of adaptation level theory can account for the category boundary shifts found in both adaptation and anchoring experiments Thus, at this point, three possible explanations for the previous vowel adaptation results have been outlined: (1)the fatigue of a set of feature detectors (Morse et al., 1976); (2) changes in auditory memory (Sawusch & Pisoni, Note 1) where adaptation level theory may provide a quantitative description of these changes; and (3) a change in labeling induced by a response bias where range-frequency theory may describe the results The present set of experiments was intended to explore these various alternative explanations of selective adaptation with vowels and to explore the similarities and differences between adaptation and anchoring E~ERIMENT1 The first experiment was undertaken to extend the anchoring procedure of Sawusch and Pisoni (Note 1) to the full 13 vowel series used by Morse et al (1976) In this procedure, one vowel occurs more often than each of the other vowels If these vowels are widely spaced in time and if the extra occurrences of the more frequently occurring vowel (anchor) are interspersed among the rest of the vowels, then it is very unlikely that any feature detectors would be fatigued Rather, fatigue would seem to depend upon close, repeated presentation of certain critical information These conditions not occur in the anchoring procedure as used by Sawusch and Pisoni (Note 1) Thus, if we obtain results similar to the category boundary shifts found by Morse et al (1976), using the same 13 vowels in an anchoring procedure, we will have evidence that is inconsistent with a feature detector fatigue interpretation Method Subjects The subjects were 33 undergraduates at SUNY/Buffalo who participated as part of a course requirement None of the subjects had previously participated in a speech experiment All subjects were right-handed native speakers of English with no reported history of either speech or hearing disorder Stimuli The same 13 vowel set used by Morse et al (1976) was used in this experiment These vowels were originally generated by Pisoni (Note 4) on the vocal tract analogue synthesizer at the Research Laboratory of Electronics, Massachusetts Institute of Technology These stimuli varied in their formant frequencies for their first three formants Formant frequencies for this series are shown in Table All of the vowels were steady state (no change in formant frequencies over time) and 300 msec in duration A complete description of these vowel stimuli can be found in Pisoni (Note 4) These 13 vowels were recorded on audiotape and then digitized using the PDP-11 computer in the Speech Perception Laboratory at Indiana University These stimuli were then reconverted into analogue form and recorded to make eight test tapes Two of these tapes were control tapes On each of these tapes, the 13 stimuli occurred five times each in random order The two [i] anchor tapes each contained 40 presentations of Stimulus ([i]) and of each of the other 12 stimuli in random order The two [I] and two [~ ] anchor tapes were contructed in a similar manner with 40 occurrences ofStimuins ([I])or Stimulus 13 ([~ ]) and of each of the other 12 stimuli in random order However, the order of stimuli was constrained such that no single stimulus ever occurred more than three times in succession All tapes were recorded with sec between stimuli Procedure The subjects were divided into three groups of 11 subjects each They were run in small groups of from four to seven 296 SAWUSCH AND NUSBAUM Table First, Second, and Third Formant Frequencies for the Vowel Stimuli Stimulus F1 F2 F3 10 11 12 13 270 285 298 315 336 353 374 397 420 444 472 501 530 2300 2262 2226 2180 2144 2103 2070 2032 1999 1966 1926 1898 1858 3019 2960 2902 2836 2776 2719 2666 2628 2581 2556 2544 2518 2492 [t(10) = 2.06, 05 < p < 1, tbr the difference between [i]-[I] and [I]-[~1 category boundaries before and after anchoring] This exactly parallels the results of Morse et aL (1976), who also found a marginally significant shrinkage of the [I] category following adaptation with [I] (Stimulus 7) The [~] anchor results are shown in Figure Both category boundaries showed a significant shift toward the [~] category [t(10)= 2.07, p < 05, and t(10) = 3.45, p < 005, for the [i]-[I] and [I]-[~] category boundaries, respectively] The mean boundary shifts from the present experiment and the adaptation experiment of Morse et al (1976) are shown together in Table Although the absolute magnitudes of the boundary shifts differ, the patterns of results are markedly similar Note- Taken from Pisoni, (19 71) subjects at a time Each subject participated in two 1-h sessions on successive days The stimulus tapes were reproduced on a Revox A-700 tape deck and presented binaurally to subjects at an intensity of 80 dB SPL (for the Stimulus vowel [i]) via Telephonies TDH-39 matched and calibrated headphones Each group fistened to the two control tapes at the beginning of each session The subjects were informed that they would be listening to synthetic vowels that: would sound like the [i] in beet, the [I] in bit, or the [~] in bet They were asked to identify each vowel as [i], [I], or [~] and write down their responses in prepared booklets The subjects were also asked to give a rating response after each identification response which indicated how sure they were that they had identified the vowel correctly A 4-point scale was used with a indicating that the subject was positive his (her) identification was correct, a indicating a probable correct, a indicating a possible correct, and a indicaring a guess Following the control tapes, each of the three groups listened to a different set of anchor tapes (the same anchor on both days) The subjects were not given any new instructions concerning these tapes They used the same three category-plus-rating response set By the end of the experiment, each subject had provided at least 20 responses to each of the 13 stimuli under both the control and one of the anchor conditions Results The identification plus rating responses were converted into ,an 8-point rating scale, with a rating of indicating an extremely confident response for a particular category and an indicating extreme confidence that the syllable was not in the particular category The results for the [i] anchor group using this scale are shown in Figure Both the [i]-[I] category boundary and the [I]-[~] category boundary showed significant shifts toward [i] [tOO) = 9.48, p < 001, for the [i]-[I] boundary; t(10) = 3.05, p < 01, for the [I]-[ ~ ] boundary], t The results for the [I] anchor group are shown in Figure No significant shift in the [i]-[I] category boundary was found It(10) = 32, p >.25] A significant shift in the [I]-[~ ] category boundary toward [I] was found [t(10) : 4.00, p < 005] The overall tendency in subjects’ responses was for the [I] category to shrink This shrinkage was marginally significant [i] Anchor Group ~7 -Ix ~< Ix Ld STIMULUS 11 13 Figure Baseline (solid) and [i]-anchored (dashed) rating functions for the [i] anchor group: [i] (circles), [I] (triangles), and [~l (squares) rating functions are shown separately ([i] ratings for Stimuli 11-13 and Id ratings for Stimuli 1-3 omitted) [I] Anchor Group 6" 5I Ill STIMULUS 11 13 Figure Baseline (solid) and Ill-anchored (dashed) rating functions for the [II anchor group: [i] (circles), Ill (triangles), and [~] (squares) rating functions are shown separately ([i] ratings from Stimuli 11-13 and [~] ratings from Stimuli 1-3 omitted) CONTEXTUAL EFFECTS IN VOWEL PERCEPTION Anchor Group 297 counted with it, there were still no substantial runs (greater than seven occurrences) of "equivalent" stimuli in these test orders, and even these runs always maintained the sec between stimuli spacing Thus, we feel that feature detector fatigue is at best a highly improbable explanation of these vowel anchoring resuits Ill This leaves two alternative explanations for the anchoring and adaptation effects found with vowels: changes in auditory memory (auditory contrast) or changes in labeling (response bias) However, before proceeding with a test of these two possibilities, a couple of differences between these results and those 13 of Morse et al (1976) and Sawuseh and Pisoni STIMULUS (Note 1) should be noted First, Morse et al (1976) found no adaptation effect on either the [i]-[I] cateFigure 4, Baseline (solid) and [~]-anchored (dashed) rating functions for the [~] anchor group: [il (circles), [11 (triangles), and gory boundary or the [I]-[~] boundary using the [I] [~l (squares) rating functions are shown separately ([i] ratings for adaptor In the present experiment, the [I] anchor did Stimuli 11-13 and [~| ratings for Stimuli 1-3 omitted) yield a significant shift in the [I]-[,] category boundary toward [I] However, Morse et al did find a Discussion trend for the [I]-[~] boundary to move toward [I] The results of Experiment and Morse et al that was not significant (see Table 2) In addition, a (1976) indicate that anchoring and adaptation with marginally significant shrinkage in the [I] category vowels produce similar patterns of results Although following [I] adaptation/anchoring was found in both the overall shifts in category boundaries were some- Morse et al and Experiment what smaller for anchoring, this may have been due to The second item to note is that Sawusch and Pisoni the basic difference between adaptation and anchor(Note 1) found a shift in the [i]-[I] category boundary ing procedures Morse et al (1976) presented their with an [I] anchor Both Morse et al (1976) and adaptors repeatedly for (100 presentations) for Experiment I found no such shift In the experiments every set of 13 test stimuli In the present anchoring of Sawusch and Pisoni (Note 1), only the first vowels procedure, 35 extra presentations of the anchor were of the 13 vowel series were used and the [i]-[I] category mixed in with a set of 65 test stimuli This difference boundary in the equiprobable control condition was in number of adaptors vs anchors may account for near the midpoint of their series (4.1 stimulus units) the overall smaller shifts in the present experiment By comparison, the [i]-[I] category boundaries in Thus, it seems reasonable to conclude that adaptation and anchoring affect the same level(s) of vowel pro- Morse et al (1976) and Experiment were at 6.15 and 5.4 stimulus units, respectively Thus, the [i]-[I] cessing and have similar effects at this level (or levels) category boundaries of Morse et al and Experiment This equivalence between anchoring and adaptation may have already been "shifted" into the [I] category, procedures runs counter to the feature detector faand further exposure to an [I] adaptor/anchor protigue explanation offered by Morse et al (1976) The duced no further shift wide spacing between extra occurrences of the anchoring vowel makes fatigue of feature detectors (relative EXPERIMENT or otherwise) quite unlikely Recently, Simon (Note 5) has reported the results of If the explanation outlined above for the failure to an adaptation experiment in which the interval be- find an [i]-[I] category boundary shift with an [I] tween adaptor repetitions was varied Her results adaptor/anchor is correct, then, using only the first indicate significant category boundary shifts with seven vowels of the test series, we should find an [I] 1,750 msec between stimuli (using 16 repetitions of the adaptor before each block of test trials) If one Table accepts the feature detector fatigue explanation, then Compaxison of Magnitudes of Category Boundary these results might lead one to expect that fatigue may Shifts for Anchoring and Adaptation still occur with sec between stimuli However, [i] Adaptor/ It] Adaptor/ [e] Adaptor/ Simon’s results also show that the optimal adaptor Anchor Anchor Anchor repetition interval was around 750 msec, which is sub[il-[II [If-[el Ill-Ill [I]-[~] [i]-[I] [I]-[~] stantially less than sec Secondly, in Experiment 1, the extra presentations of the anchor were widely Adaptation 1.18 1.02 15 60 96 1.19 interspersed with other stimuli Even if other stimuli Anchoring 1.52 53 06 47 37 78 very similar to the anchor (within stimulus units) are Note Adaptation data taken from Morse et al, (1976) 298 SAWUSCH AND NUSBAUM anchor shift (as per Sawusch & Pisoni, Note 1) Experiment was conducted in part to check this explanation The other rationale for Experiment was to investigate one aspect of a response bias explanation of anchofing In particular, the number of response categories available to the subject should influence the magnitude of change in the category boundary when the range of stimuli is extended (cf Parducci, 1975) In particular, if an [r] anchor is added to an [i]-[I] test series, subjects should show a larger movement of the category boundary toward [I] if they only have two response categories ([i] and [I]) than if they have three response categories ([i], [I], and [~]) This result would occur because, in one case, subjects would be forced to stretch two response categories to cover an increased range of vowel stimuli (cf range-frequency theory, Parducci, 1965) If this pattern of results is found, then we will have shown that it is possible to induce a labeling change/response bias during vowel identification in an anchoring paradigm to simply check them off when they occurred These two different [r ] anchor groups represent the manipulation of the number of response categories available to subjects Results The two-response category plus 4-point rating were unfolded into an 8-point rating scale with a representing a positive [i] and an representing a positive [I] The results for the [i] and [I] anchor groups are shown in Figure on the left and fight sides, respectively Both anchoring conditions produced contrast effects For both groups, a significant shift in the category boundary toward the anchor (relative to the equiprobable control) was found [t(9) = 5.42, p < 001, for the [i] anchor; t(9) = 4.00, p < 01, for the [I] anchor] The average boundary shifts were 77 stimulus units toward [i] for the [i] anchor group and 42 units toward [I] for the [I] anchor group These results replicate the earlier findings of Sawusch and Pisoni (Note 1) The results of the two [~] anchor groups are shown in Figure On the left is the group which was informed about the 40 added [~] vowels A significant Method Subjects The subjects in this experiment were 40 undergraduates shift in the category boundary was found [t(9) = 6.89, p < 001] The magnitude of this shift was at the State University of New York at Buffalo, who participated as part of a course requirement All were right-handed native almost double that found with the [I] anchor (.81 vs speakers of English with no reported history of either speech or 42 units) and was significantly greater than the [I] hearing disorder anchor shift [t(18) = 2.42, p < 05, for a two-tailed, Stimuli The stimuli in this experiment were taken from the same independent t test] On the fight are the results for set of 13 synthetic steady-state vowels that were used in Experiment These stimuli were converted to analog form and recorded the [~] anchor group, which was not informed about to make tbur test tapes the addition of the new [~] vowel in the anchor tapes The control tape contained 10 occurrences of each of the first (labeled [~] as [I]) This group also showed a signifseven stimuli in random order The [i] anchor tape contained 40 icant movement of the category boundary [t(9) = occurrences of Stimulus ([i]) and 10 occurrences of each of the other six stimuli (Stimuli 2-7) Similarly, the [I] anchor tape con- 8.21, p < 001] The mean shift of 1.78 stimulus units tained 40 occurrences of Stimulus and 10 each of Stimuli 1-6 The was significantly larger than both the [I] anchor group [r ] anchor tape contained 40 occurrences of Stimulus 13 ([~ ]) and shift and the other [~] anchor group shift [t(18) = 10 each of Stimuli 1-7 in random order In all four tapes, the 5.62, p < 001, and t(18) = 3.95, p < 002, respecorder of stimuli was constrained such that no one stimulus e~er tively, for the two-tailed independent t tests] Of the occurred more than three times in succession All tapes were re10 subjects in the [~]-as-[I] anchor group, exhibcorded with sec between stimuli Procedure Subjects were divided into four groups of 10 each They were run in smaller groups of from four to six subjects at a time The audiotapes were reproduced and presented to the subjects in the same manner as in Experiment All subjects were run lbr h At the beginning of their session, all subjects listened to two presentations of the control tape The subjects were informed that they would be listening to synthetic examples of the vowels [i] as in beet and Ill as in bit and that they were to identify each vowel as either [i] or [I] and write down their responses in prepared response booklets The subjects were also asked to give a rating response indicating how sure they were that they had identified each vowel correctly The same 4-point scale used in Experiment was used here Following the two control tape presentations, each group heard two presentations of an anchoring tape The [i]-, [I]-, and [r Ianchor tapes were each given to one group These subjects were not given any new instructions about these tapes They were asked to continue to use the two-category ([i] and [I]) and 4-rating-point response This [r ] anchor group will be referred to as the [r ] as [I] group The fourth group also listened to the [E ] anchor tape However, these subjects were told that 40 occurrences of an lr] vowel were going to be added Further, the occurrences of the [~] vowel were premarked on their response sheets The subjects were asked Anchor [I] Anchor STIMULUS VALUE Figure Baseline (solid) and anchored (open-dashed) rating functions for the [il anchor (left) and 111 anchor (righl) groups in Experiment CONTEXTUAL EFFECTS IN VOWEL PERCEPTION Anchor Labeled as [I] Anchor 299 within limits, lead to large changes in identification Thus, part of the anchoring and adaptation effects observed with vowels may be due to a change in labeling induced by a response bias EXPERIMENT [i] J/,d,/’ [H 123 STIMULUS VALUE 456 Figure Baseline (solid) and anchored (open-dashed) rating functions for the three-response category [~] anchor group (left) and the two-response category [~] as [1| anchor group (right) in Experiment ited a rather drastic change in their identification of the seven test vowels These four subjects identified all seven of the test vowels on the [~]-as-[I] anchor tape as [i] This result made it impossible to determine a precise category boundary for these subjects Consequently, Stimulus was taken as the location of the category boundary for these four subjects Each of the other subjects showed a distinct category boundary However, all 10 of these subjects showed decreases in their ratings for Stimuli through Discussion The results of the [i] and [I] anchor groups replicate Sawusch and Pisoni (Note 1) These results demonstrate rather conclusively that an [I] anchor can cause a shift in the [i]-[I] category boundary It should be noted that the [I]-anchored and the premarked [~ ]-anchored category boundaries were at 4.9 and 5.3 stimulus units, respectively Both of these values are less (nearer the [i] end of the series) than the equiprobable [i]-[I] category boundary in Experiment (5.4 stimulus units) Thus, there seem to be limited regions within which shifts in category boundaries under anchoring (and adaptation) with vowels can be obtained In Experiment 1, the [i]-[I] category boundary was apparently near one such limit and no shift was found for an [I] anchor The other primary result from Experiment is the difference between the two [~] anchor groups The [~] as [I] group, which was forced to use two responses ([i] and [I]) to cover all stimuli, showed a much larger shift in their category boundary than the premarked (three-response category) [~] group These results are consistent with a response bias explanation of vowel anchoring such as range-frequency theory (Parducci, 1963, 1965, 1975), and indicate that discrepancies between the subject’s internally available response categories and those that are externally imposed can, At this point, a feature detector fatigue explanation of anchoring and adaptation effects with vowels seems to be untenable This leaves at least two potential explanations: changes in auditory memory and response bias The results of Experiment imply response bias in one particular case, where the response categories available to the subject are manipulated However, this still leaves us with the question of what is happening in the other anchoring conditions, where appropriate response categories are available In order to distinguish between changes in auditory memory and response bias, relatively specific predictions from these two explanations are needed In particular, a more specific response bias model is needed One relatively simple form of response bias explanation, a probability matching type of explanation, can be tested directly If the subject is simply trying to use the two (or three) vowel response categories equally often, then the presence of extra members of one category (an anchor) would cause’ the subjects to shift their category boundary to maintain equal response frequencies If this were the case, then knowledge of the extra occurrences of one stimulus might be expected to diminish the contrast effect found However, there was no appropriate control to test this explanation in Experiments or Furthermore, the number of response categories did have a large effect in Experiment This response-categories effect may have been due to a probability matching effect brought about by subjects treating the [~] anchor as an [I] vowel in the [~] as [I] anchor group Experiiment was intended as a direct test of the probability matching hypothesis, using a situation in which the number of response categories (and range of vowels) was held constant and only instructions were varied If instructions as to the identity and frequency of the anchor have no effect on anchoring, then the relatively simple probability matching model can be ruled out On the other hand, since the actual vowels presented to the subjects are identical under the two instructional conditions, the auditory-memory explanation predicts that the same anchoring effects should be found for the different sets of instructions Method Subjects The subjects were 32 undergraduates at SUNY/ Buffalo, who participated as part of a course requirement None of these subjects had participated previously in a speech experiment, and all met the same requirements as previous subjects Stlmnli The first seven vowels ([i]-[I]) of the 13-vowel set used previously were employed They were recorded on audiotape to 300 SAWUSCH AND NUSBAUM make a control tape, an [i] (Stimulus 1) anchor tape, and an [i] Anchor (Stimulus 7) anchor tape, as per Experiment The anchoring Old Instruct[ors vowel occurred four times as often as each of the other vowels Procedure The subjects were divided into four groups of eight subjects each They were run in small groups of from three to five {D subjects at a time for one 1-h session The audiotapes were presented to the subjects as in Experiments and At the beginning of each session, all subjects listened to two presentations of the control tape The subjects used the same two-category-plus-4-pointrating response procedure that was used in Experiment Following the control tapes, each group listened to two presentations of one of the anchor tapes Two of the groups listened to the [i] anchor tape, and the other two listened to the [I] anchor tape Of the two [i] anchor groups, one received the same instructions given to the [i] anchor subjects in Experiments and When the [i] anchor 234567 1234567 tapes were presented, they were told only that the tape contained 100 [i] and [I] vowels for them to respond to The other [i] anchor STIMULUS VALUE group received new instructions They were told that the difference Figure Baseline (solid) and anchored (open-dashed) rating between the previous two (70-item) tapes and the next two (100functions for the two [1] anchor groups in Experiment 3: old item) tapes was that an extra 30 occurrences of the [i] vowel were instructions on the left and new instructions on the right randomly distributed throughout each of the next (100-item) tapes The two [I] anchor groups were similarly given old and new instructions In all other respects, except instructions, the two [i] anchor icantly different [t(14)= 11, p >.80, for a twogroups (It] anchor groups) were treated identically tailed, independent t test] Results The identification-plus-rating responses were converted into an 8-point rating scale (as per Experiment 2) with a being a positive [i] and an a positive [I] The results for the two [i] anchor groups are shown in Figure In both cases, a significant shift in the category boundary toward the anchoring stimulus was found [t(7) := 3.73, p < 01, for the old instructions; t(7) = 6.14, p < 001, for the new instructions] The mean shifts of 51 stimulus units for the old instructions and 66 for the new instructions were not significantly different It(14)= 90, p >.20, for a two-tailed, independent t test] The results for the two [I] anchor groups are shown in Figure Again, a significant category boundary shift toward the [I] anchor was found for both old and new instructions groups It(7) = 5.21, p < 005, and t(7) = 3.80, p